Copertina

Overconfident Errors Need Stronger Correction: Asymmetric Confidence Penalties for Reinforcement Learning

Fuente: arXiv

Salvato in:

Dettagli Bibliografici
Autori principali:	Xu, Yuanda, Sang, Hejian, Zhou, Zhengze, He, Ran, Wang, Zhipeng
Natura:	Preprint
Pubblicazione:	2026
Soggetti:	Machine Learning Artificial Intelligence
Accesso online:	Acceder al recurso
Tags:	Aggiungi Tag Nessun Tag, puoi essere il primo ad aggiungerne!!

Documenti analoghi

PACED: Distillation and On-Policy Self-Distillation at the Frontier of Student Competence
di: Xu, Yuanda, et al.
Pubblicazione: (2026)

Not all tokens are needed(NAT): token efficient reinforcement learning
di: Sang, Hejian, et al.
Pubblicazione: (2026)

Beyond GRPO and On-Policy Distillation: An Empirical Sparse-to-Dense Reward Principle for Language-Model Post-Training
di: Xu, Yuanda, et al.
Pubblicazione: (2026)

TIP: Token Importance in On-Policy Distillation
di: Xu, Yuanda, et al.
Pubblicazione: (2026)

CRISP: Compressed Reasoning via Iterative Self-Policy Distillation
di: Sang, Hejian, et al.
Pubblicazione: (2026)

Robust Batch-Level Query Routing for Large Language Models under Cost and Capacity Constraints
di: Markovic-Voronov, Jelena, et al.
Pubblicazione: (2026)

Debunk the Myth of SFT Generalization
di: Lin, Xiaofeng, et al.
Pubblicazione: (2025)

Aligning Diffusion Language Models via Unpaired Preference Optimization
di: Jindal, Vaibhav, et al.
Pubblicazione: (2025)

Just Enough Thinking: Efficient Reasoning with Adaptive Length Penalties Reinforcement Learning
di: Xiang, Violet, et al.
Pubblicazione: (2025)

Training with Confidence: Catching Silent Errors in Deep Learning Training with Automated Proactive Checks
di: Jiang, Yuxuan, et al.
Pubblicazione: (2025)

LLMs are Overconfident: Evaluating Confidence Interval Calibration with FermiEval
di: Epstein, Elliot L., et al.
Pubblicazione: (2025)

Scaling In-Context Online Learning Capability of LLMs via Cross-Episode Meta-RL
di: Lin, Xiaofeng, et al.
Pubblicazione: (2026)

Agentic Uncertainty Reveals Agentic Overconfidence
di: Kaddour, Jean, et al.
Pubblicazione: (2026)

Sparse Training for Federated Learning with Regularized Error Correction
di: Greidi, Ran, et al.
Pubblicazione: (2023)

Beyond Correctness: Confidence-Aware Reward Modeling for Enhancing Large Language Model Reasoning
di: He, Qianxi, et al.
Pubblicazione: (2025)

Uncalibrated Reasoning: GRPO Induces Overconfidence for Stochastic Outcomes
di: Bereket, Michael, et al.
Pubblicazione: (2025)

Confident or Seek Stronger: Exploring Uncertainty-Based On-device LLM Routing From Benchmarking to Generalization
di: Chuang, Yu-Neng, et al.
Pubblicazione: (2025)

Single-Trajectory Distributionally Robust Reinforcement Learning
di: Liang, Zhipeng, et al.
Pubblicazione: (2023)

Mitigating Overconfidence in Out-of-Distribution Detection by Capturing Extreme Activations
di: Azizmalayeri, Mohammad, et al.
Pubblicazione: (2024)

OffSeeker: Online Reinforcement Learning Is Not All You Need for Deep Research Agents
di: Zhou, Yuhang, et al.
Pubblicazione: (2026)

Reinforcement Learning for Control with Probabilistic Stability Guarantee: A Finite-Sample Approach
di: Han, Minghao, et al.
Pubblicazione: (2026)

Meta-Cognitive Reinforcement Learning with Self-Doubt and Recovery
di: Zhang, Zhipeng, et al.
Pubblicazione: (2026)

LLMBoost: Make Large Language Models Stronger with Boosting
di: Chen, Zehao, et al.
Pubblicazione: (2025)

ECHO: Entropy-Confidence Hybrid Optimization for Test-Time Reinforcement Learning
di: Zhao, Chu, et al.
Pubblicazione: (2026)

On Stronger Computational Separations Between Multimodal and Unimodal Machine Learning
di: Karchmer, Ari
Pubblicazione: (2024)

Stochastic Penalty-Barrier Methods for Constrained Machine Learning
di: Bosák, Adam, et al.
Pubblicazione: (2026)

What If Consensus Lies? Selective-Complementary Reinforcement Learning at Test Time
di: Yan, Dong, et al.
Pubblicazione: (2026)

CAdam: Confidence-Based Optimization for Online Learning
di: Wang, Shaowen, et al.
Pubblicazione: (2024)

Approximate Subgraph Matching with Neural Graph Representations and Reinforcement Learning
di: Li, Kaiyang, et al.
Pubblicazione: (2026)

Response-Level Rewards Are All You Need for Online Reinforcement Learning in LLMs: A Mathematical Perspective
di: He, Shenghua, et al.
Pubblicazione: (2025)

Bounding-Box Inference for Error-Aware Model-Based Reinforcement Learning
di: Talvitie, Erin J., et al.
Pubblicazione: (2024)

Confidence Is All You Need for MI Attacks
di: Sinha, Abhishek, et al.
Pubblicazione: (2023)

Decoupling Reasoning and Confidence: Resurrecting Calibration in Reinforcement Learning from Verifiable Rewards
di: Ma, Zhengzhao, et al.
Pubblicazione: (2026)

Towards Differentially Private Reinforcement Learning with General Function Approximation
di: He, Yi, et al.
Pubblicazione: (2026)

Learning to Clean: Reinforcement Learning for Noisy Label Correction
di: Heidari, Marzi, et al.
Pubblicazione: (2025)

Adviser-Actor-Critic: Eliminating Steady-State Error in Reinforcement Learning Control
di: Chen, Donghe, et al.
Pubblicazione: (2025)

A Survey of Explainable Reinforcement Learning: Targets, Methods and Needs
di: Saulières, Léo
Pubblicazione: (2025)

Reinforcement Learning in Dynamic Treatment Regimes Needs Critical Reexamination
di: Luo, Zhiyao, et al.
Pubblicazione: (2024)

Value Bonuses using Ensemble Errors for Exploration in Reinforcement Learning
di: Wahab, Abdul, et al.
Pubblicazione: (2026)

Deep Reinforcement Learning and The Tale of Two Temporal Difference Errors
di: Rojas, Juan Sebastian, et al.
Pubblicazione: (2026)